US letter - SecurityMatters

Monitoring Industrial Control Systems
to improve operations and security
An overview of the threats to Industrial Control Systems and
the technologies to protect them
Authors
Dr.
Dr.
Dr.
Dr.
Dr.
Sandro Etalle, [email protected]
Cliord Gregory, [email protected]
Damiano Bolzoni, [email protected]
Emmanuele Zambon, [email protected]
Daniel Trivellato, [email protected]
01 December 2013
Version 1.0
Contents
1 The need for monitoring
2
2 Threat and vulnerability analysis
5
1.1 Assessing the Risk . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.1 Threat sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.2 The weak spots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.3 Example threats . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3 Mitigation solutions
3.1
3.2
3.3
3.4
Signature-based . . . .
Rule-based . . . . . . .
Non-signature based .
Technology comparison
4 Conclusions
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
4.1 Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.2 About SecurityMatters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.3 About the Authors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3
5
6
8
10
10
11
12
14
15
16
16
17
1 The need for monitoring
Industrial Control Systems (ICS) sit at the core of every industrial process - from power generation to water treatment and manufacturing. The term ICS refers to the set of devices that
govern the process to guarantee safe and successful execution, and include Supervisory Control and Data Acquisition (SCADA) systems, Distributed Control Systems (DCS), and other
control systems such as Remote Terminal Units (RTU) and Programmable Logic Controllers
(PLC). A malfunction in any of these systems might cause the entire industrial process to fail,
with serious consequences in terms of economic loss and in compromised public safety. For
instance, consider disruption of an electricity transmission network; an incorrect distribution
of power might aect availability to households, oces, hospitals, etc. Similarly, a faulty
component that regulates the amount of chemical substances in a pharmaceutical production
process might lead to entire batches of harmful compounds.
The need to monitor ICS and more generally industrial networks is advocated in many
venues and has been included in several recommendations, guidelines, and standards, such
as the Electricity Subsector Cybersecurity Capability Maturity Model (ES-C2M2); the North
American Electric Reliability Corporation Critical Infrastructure Protection (NERC CIP); and
the still in progress M/490 Directive from the European Union. There are several reasons
why monitoring should be an integral part of operations, and may even be considered a competitive advantage. The marked increase of cyberattacks directed against ICS, as frequently
reported by the ICS Cyber Emergency Response Team (ICS-CERT) of the U.S. Department
of Homeland Security, is only one of them. Stuxnet, Duqu, Flame, and Gauss have seized all
the attention of the media, but represent only a small portion of the cyberthreats targeting
ICS.
Today, organizations rely heavily on third parties (i.e. contractors, vendors, etc.) to
streamline their operations. Smaller organizations, such as local utilities, have consultants
programming and maintaining their PLCs, while larger organizations have direct lines connecting their process control devices to vendors for 24/7 support and maintenance. In all cases
connectivity comes with marvelous gains in productivity, but adds a totally new dimension of
risks which must be mitigated to secure ICS networks. The shift in how industrial networks
are operated and maintained makes it crucial for organizations to monitor their infrastructure
not only against cyberattacks, but also against system misuse by third parties. Next to this,
c
2013 SecurityMatters B.V.
2
www.secmatters.com
there is an ever-present threat coming from insiders, including disgruntled employees, human
error, and network and system misconguration, which could aect or lead to disruption of
the production process. All these threats strongly impact the bottom line revenue of an
organization, and as such should receive attention of the boardroom daily.
Organizations can mitigate these threats by implementing a solid monitoring infrastructure for ICS and back-oce networks. With an eective monitoring infrastructure in place,
organizations are not only able detect problems at an earlier stage, but can also mitigate
consequences before any real damage is done and recover more quickly from an incident, no
matter the nature. Continuous monitoring and early analysis of identied issues help organizations pinpoint root causes of a problem and enforce eective countermeasures and remedial
actions.
Before implementing any monitoring infrastructure or even considering a specic solution,
however, organizations should assess their exposure to the risks and threats to their processes
and devices. Quite often organizations overlook this phase of the security process, and rush
into picking a solution that does not match their needs or expectations. This document is not
intended to provide details about risk assessment procedures. Nevertheless, it is important to
understand its key concepts and identify the steps that lead to the selection of appropriate
monitoring solutions.
1.1
Assessing the Risk
The security process within an organization is an unending cycle that aims at continuously
improving the security posture of the organization. In order to determine the next steps,
estimate eort and measure improvement (or regression) over time, security personnel need
to assess the organization's current security status in a number of areas.
A typical risk management process consists of three phases: risk analysis, risk evaluation,
and risk mitigation. During the risk analysis phase, an organization must identify its key assets
and their vulnerabilities, as well as the related threats and threat sources. Threat sources can
be either external to an organization, such as foreign intelligence agencies, cyberterrorists,
and hacktivists, or internal, such as poorly trained or disgruntled employees, contractors, and
vendors. It is unlikely that a foreign intelligence agency would specically target a local utility
organization, whereas a multinational company might very well be targeted by cybercrime.
Each adversary has dierent characteristics in terms of funding and capability to exploit an
organization's critical assets.
To eectively analyze risk, it is important to carefully assess impact and likelihood of each
threat being exercised against an asset's specic vulnerability. While assessing the likelihood
of a threat is generally easier in most IT environments, it is nearly impossible in ICS networks,
due to the scarce historical data available regarding past incidents. The best way available
to proceed is thus to take into consideration high-impact threats and determine the cost and
eort needed to recover productivity. Threats are then ranked based on their overall cost and
alleged likelihood during the risk evaluation phase. The result of this phase is a prioritized list
of threats to be mitigated.
Finally, an organization must decide \what to do" with the identied threats, i.e. whether
they can be avoided, should be accepted, or might be reduced by means of adequate controls.
In traditional risk management, there are several types of control that can be applied to reduce a risk, including: legal, deterrent, preventive, monitoring, detective, corrective, etc. The
implementation of each of these controls might involve dierent elements of an organization,
such as people (e.g. good practices), the physical world (e.g. access control), procedures
(e.g. change management, disaster recovery), and technology (e.g. antivirus, intrusion detection/prevention systems). The selection of the right (combination of) controls should
aim at minimizing the likelihood and impact of the identied threats, thereby maximizing the
benet for the organization.
c
2013 SecurityMatters B.V.
3
www.secmatters.com
In the rest of the whitepaper we discuss the most prominent cyberthreats to ICS and their
possible impact on an organization; we also give an overview of the existing technologies that
can be employed to mitigate them.
c
2013 SecurityMatters B.V.
4
www.secmatters.com
2 Threat and vulnerability
analysis
A threat and vulnerability analysis is a mandatory step that should precede all security-related
decisions, including decisions regarding which procedures to adopt and which prevention and
monitoring solutions to employ. Unfortunately, more often than not, this step is skipped
altogether.
In this section we analyze the possible threats to ICS. The threat analysis is performed
in three steps: rst, we discuss the major sources of threat to ICS (adversaries) and their
capabilities; then, we characterize ICS networks, pinpointing the key components and their
vulnerabilities; lastly, we present a number of example methods that adversaries can use to
disrupt or sabotage the victim's industrial process. The identication and selection of the
most prominent threats to an organization is instrumental in determining the most appropriate
countermeasure to be employed for mitigation.
2.1
Threat sources
Threats to ICS can come from several sources, including hostile governments, terrorist groups,
competitors, malicious intruders, and disgruntled or careless employees [4, 7]. These adversaries can be classied in many ways. For simplicity, here we consider the following two
categories:
Outsiders: adversaries who are not associated with the organization and acting from
the outmost perimeter.
Insiders: adversaries who are already within the target organization's perimeter.
Outsiders include governments and foreign intelligence services, hackers and hacktivists, industrial spies, (cyber-) terrorists, and organized crime. Insiders include disgruntled employees,
careless or poorly trained employees, contractors and vendors.
c
2013 SecurityMatters B.V.
5
www.secmatters.com
Outsiders have varying degree of knowledge and motives. They may perform attacks for
information gathering, espionage activities, or to disrupt or weaken the target's processes.
They are usually motivated by political, monetary, or reputation reasons.
While outsiders typically have higher funds and resources, the insider threat should not be
underestimated by organizations. Joe Weiss writes in his book \Protecting ICS from electronic
threat" [8] that insiders might be the biggest concern nowadays, due to the specic knowledge
of and access to the organization's infrastructure. The NIST Special Publication 800-82 [7]
provides an accurate description of the insider threat for ICS: \The disgruntled insider is a
principal source of computer crime. Insiders may not need a great deal of knowledge about
computer intrusions because their knowledge of a target system often allows them to gain
unrestricted access to cause damage to the system or to steal system data. The insider
threat also includes outsourcing vendors as well as employees who accidentally introduce
malware into systems. [. . . ]Impacts have ranged from trivial to signicant damage to the
ICS and eld devices. Unintentional impacts from insiders are some of the highest probability
occurrences." In the rest of the section we describe the dierent types of insiders, their
capabilities and intent.
Disgruntled employees This adversary includes both current and former employees that
are upset or dissatised with the organization. Their motivation can also be fame, greed,
divided loyalty, delusion or monetary gain [5]. Depending on their level of access to the
infrastructure, disgruntles employees can pose a serious threat to the employer. Disgruntled
current employees will have legitimate credentials and possibly unrestricted access to the
organization's infrastructure. Former employees may still have valid credentials, either because
the old access rights have not been revoked, or because the employee secretly created new
access credentials before the termination of employment. All current and former employees
have process knowledge and may be aware of system vulnerabilities.
Careless or poorly trained employees and unintentional mistakes Employees with a lack
of training, concern or attentiveness can pose a serious threat to their organization. This
adversary does not have the intention of harming the organization or to cause incidents. In
most cases the misuse stems from unintentional mistakes, whose eects on normal operations
could remain unnoticed for a long period of time. For example, employees may unknowingly
introduce malware to the internal networks by using an infected private storage media or
laptop.
Contractors and vendors Contractors and vendors are third parties used by organizations
for carrying out specic tasks and providing specialized services. Organizations often depend
on the expertise and services from such third parties. Vendors often require remote access
to oer additional (maintenance) services, such as updates and diagnostics. While organizations usually limit and control the access rights to the necessary minimum, once a remote
connection to eld equipment is established any misuse could take place.
2.2
The weak spots
Adversaries aiming at disrupting an organization's production processes will attempt to access,
damage, or improperly use its critical assets. An asset is broadly dened as \anything, which
has value to an organization, its business operations and its continuity " [1]. The ISA 95
standard denes a multi-level model for industrial activities. Each level includes specic
assets, provides specialized functions and has characteristic response times.
1
1 www.isa.org/ISA95
c
2013 SecurityMatters B.V.
6
www.secmatters.com
Level 0: refers to the actual physical production process. It involves the machinery used
for the process.
Level 1: denes the activities involved in sensing and manipulating the production
process. Includes RTUs, PLCs, and eld devices such as sensors, actuators, and other
I/O devices.
Level 2: involves supervising, monitoring and governing the physical processes. The
systems acting on this level are real-time controls and software, SCADA/DCS servers,
and operator workstations (HMIs).
Level 3: includes activities for managing production workow. The collection of systems
acting on level 3 can be referred to as Manufacturing Operations Management Systems
(MOMS).
Level 4: focuses on managing the business-related activities of the manufacturing operations. Enterprise Resource Planning (ERP) is the primary system. Other systems
acting on this level are Product Lifecycle Management (PLM), Customer Relationship
Management (CRM), and Human Resource Management (HRM).
Industrial networks are typically segmented to reect this classication. In particular,
activities on level 4 and 3 are carried out in corporate (or back-oce) networks, while activities
on level 2 and 1 are carried out in Supervisory and Process Control Networks respectively. A
simplied example of industrial network and its components is shown in Figure 2.1.
Figure 2.1: Example of industrial network
The most critical assets for an organization operating industrial processes are those located
in Supervisory and Process Control Networks, since they provide full control of the victim's
production process. Adversaries will attempt to take control of these assets by exploiting one
of their vulnerabilities. A vulnerability is a weakness in a component's design, implementation,
or operation and management that if exploited can lead to a security breach or an incorrect
behavior of the component. Vulnerabilities often result from programming errors such as lack
of input validation and memory safety violations. Whereas until a few years ago vulnerabilities
and exploits were available almost exclusively for Information Technology (IT) components,
i.e. Windows machines and Windows applications, in the last years this trend has inverted
its course. Today, vulnerabilities and exploits for Operational Technology (OT) components
can be purchased on an open and thriving Internet black market that grows daily.
c
2013 SecurityMatters B.V.
7
www.secmatters.com
Figure 2.2 shows the increase in the number of known vulnerabilities for ICS over the
last years [2]. These vulnerabilities aect several OT components, with SCADA server and
HMI being the most targeted ones. Although these components are essential for a company's
production process, they are becoming easier to approach and compromise from both external
and internal adversaries.
Figure 2.2: Number of known vulnerabilities for ICS
It is important to notice that the numbers in the graph report only the vulnerabilities that
have been discovered and published during the indicated year, and as such they are just the
tip of the iceberg. Often, vendors prefer not to disclose vulnerabilities of their components
for both safety and commercial reasons.
2.3
Example threats
In this section we present a number of threat scenarios involving dierent types of adversaries
introduced in Section 2.1. In particular, we present the possible eects on industrial networks
of:
1. Malicious/careless operators
2. An unexpected conguration shift
3. A targeted attack exploiting an unknown vulnerability
4. Spread of malware
Malicious/careless operators Operators and engineers must by the nature of their jobs
have a thorough knowledge of all aspects of an industrial process, and are generally holders
of \privileged access" to the network resources. In the normal process of their jobs they
may modify a system's regular workow, and they have access to sensitive information that
could be passed on to third parties for any number of reasons. Most ICS have the potential
of implementing access control mechanisms, but organizations seldom use these controls in
practice, both as a measure to save on costs of their re-conguration and to achieve higher
productivity and lower downtime when upgrading or updating their network. An employee may
disrupt the production process by issuing unusual or abnormal commands, using parameter
values that cause PLCs and eld devices to crash. A trusted employee may become a threat
due to nancial gain or dissatisfaction, or simply due to a mistake. Whatever the motivation,
the insider can cause catastrophic damage to critical infrastructure, and some have in the
past.
An unexpected conguration shift Threats to ICS do not necessarily originate from adversaries, but might arise during the normal course of operations. In fact, network and system
congurations tend to change over time, for instance due to on-site maintenance and hardware/software updates. Any organization dealing with critical processes tests network and
c
2013 SecurityMatters B.V.
8
www.secmatters.com
systems' conguration prior to deployment, and stores a copy (a blueprint) for backup in case
of disruption, so that restoring can be as quick as possible. Blueprints should be regularly
updated to reect the changes in running congurations. If for any reason blueprints are not
kept up-to-date, a restore procedure could not be eective and cause additional delays and
prolong downtime caused by an incident.
A targeted attack exploiting an unknown vulnerability ICS are not generally targeted by
what has come to be considered mainstream hackers. The ICS attacker is often motivated,
well-funded and may even be a state-sponsored organization, as research has shown for the
latest successful ICS intrusions. Attacks to ICS are carefully planned and executed employing
specically developed malware, which circumvents standard security solutions by exploiting
previously unknown vulnerabilities in target systems. The aim might be to steal nancial or
exploration data (as in the case of the Flame malware) or to disrupt operation of the provided
service (e.g. Stuxnet). These attacks may be part of a larger cyber-warfare strategy.
Spread of malware One of the main threats to ICS is malware that spreads from one
computer to the other, multiplying the damage. The harm resulting from malware spreading
around in a corporate network can be huge, and it could become unbearable if the infection
manages to reach the Supervisory or Process Control Network. Despite strict network segmentation and corporate policies, Supervisory and Process Control Networks are never truly
isolated from the external world. Security experts have pointed out that the use of air gaps (or
system isolation) is a \myth". External consultants and support and maintenance personnel
are required from time to time to transfer data to and from those networks, for instance with
USB sticks, and to carry out system software updates, thwarting network segmentation and
isolation policies. Once one of the systems is infected, malware can then spread easily and
often unnoticed into sensitive, high-security environments and force long periods of system
downtime. A real-world example of this scenario is represented by the Shamoon malware,
which hit 30.000 workstations at Saudi Aramco requiring a lengthy restore process.
2
2 http://www.symantec.com/connect/blogs/shamoon-attacks/
c
2013 SecurityMatters B.V.
9
www.secmatters.com
3 Mitigation solutions
After identifying the major sources of threat to ICS and illustrating some possible threat
scenarios, we discuss the controls that can be used to mitigate them. In particular, here we
focus on the various existing technologies that can be used for continuous monitoring of ICS
networks and the detection of threats.
There exist heterogeneous and complementary monitoring solutions to mitigate threats
to ICS. Each of them has pros and cons and there is no silver bullet. Depending on the output
of the risk assessment, an organization should select the set of solutions that best suits its
needs. Traditionally, monitoring solutions are divided into host- (or endpoint) and networkbased. Most organizations employ \regular" (or slightly modied) endpoint security solutions
developed for IT environments, such as antivirus and application whitelisting, to protect their
SCADA/DCS servers, HMIs and engineering workstations. There is a general consensus that
IT endpoint security nds straightforward application in the context of OT. These solutions,
however, will not suce when it comes to monitoring PLCs and RTUs, detecting insiders'
misbehavior, and identifying miscongurations. Network-based solutions become then the
only available choice for protecting some of the most critical assets for an industrial process.
In the next paragraphs we describe the existing network monitoring technologies, their pros,
and their cons.
3.1
Signature-based
Signature-based detection, or blacklisting, is the most common monitoring and detection
method. The idea behind this approach is simple and straightforward. Knowledgeable people,
like security analysts, make a list of known malicious and suspicious network messages and
devise signatures (i.e., byte patterns) to recognize them. Network trac is then compared
in real-time against these signatures; in case of a successful match, the event is reported to
the security operator.
Pros: The approach is simple, widely known among the IT security community, and very eective against mainstream threats, such as well-known malware and software exploits. Usually,
c
2013 SecurityMatters B.V.
10
www.secmatters.com
a signature-based solution can be deployed in a matter of hours. Some attention must be
paid to the tuning phase, which is required in order to suppress false warnings that might be
triggered by too strict signatures or even by the type of network trac in a certain OT network. An additional benet is the option of turning prevention mode on, eectively blocking
(known) misbehavior.
Cons: This approach can only detect previously known misbehavior. Skilled attackers can
easily circumvent detection, even by just morphing known attack payloads. Every time a
new misbehavior or attack is discovered, a corresponding signature must be developed and
distributed; it can take weeks before all monitoring systems are updated. In the meantime,
attack payloads that have not yet been isolated, analyzed and mapped to a signature can hit
without notice. For this reason, malware such as Stuxnet, Duqu, Flame and Shamoon spread
across systems silently for lengthy periods of time. Stuxnet leveraged four unknown software
vulnerabilities and went undetected for at least a year. Another limitation of signature-based
detection is that \benign yet incorrect" network operations will also go unnoticed, unless a
specic signature (or check) for it has been dened. At the present state of aairs, it is
impossible to have a set of signatures that provides a reasonable degree of protection for ICS
networks. More generally, blacklisting is not suitable for the ICS world, where attackers are
more motivated to remain silent, and where attacks are targeted to specic systems.
3.2
Rule-based
Rule-based monitoring, also called network whitelisting, consists of identifying and detailing
legitimate and acceptable network activities using rules, and blocking or alerting when nonmatching behavior is observed. The eectiveness and usability of rule-based solutions largely
depend on how accurate the underlying analysis of network trac is. At one extreme, there are
low accuracy solutions that work like rewalls, controlling whether network trac matches,
for example, specic combinations of IP addresses and TCP ports. At the other extreme it
is possible to devise ne-grained rules, based on deep packet inspection. Solutions featuring
deep packet inspection are capable of understanding (part of) the underlying protocol used by
the application being monitored and of blocking/detecting the invocation of certain undesired
message types (medium accuracy solutions), or even the use of eld values which do not
respect pre-determined constraints (high accuracy solutions).
The conguration of a rule-based solution is usually performed manually and involves
expert personnel that are knowledgeable of the underlying processes. This is because rules
are tailored to protect a specic environment, rather than a \general" infrastructure like
signature-based solutions do. In other words, while the same signature-based system can be
deployed on dierent networks with minor adjustments, a rule-based system requires ad-hoc
conguration for each network it is deployed on.
Pros: Rule-based solutions allow only network communications that have been explicitly
whitelisted. As a result, they oer protection from a wide range of malicious events and
behavior, including both known and new and unknown threats.
Cons: The higher the accuracy of the solution, the higher the chance that it will block an
attack, regardless of whether it is known or not. Accuracy (and thus eectiveness), however,
comes at a cost. An important drawback of rule-based solutions is constituted by their high
conguration cost, i.e., the cost of detailing all the whitelisting rules. The more accurate the
desired level of analysis, the higher the setup cost. Whitelisting all message types and message
eld values used in a production site is a daunting task that does not t most budgets or time
schedules. This can be only partly mitigated by employing a coarse-grained conguration that
fails to take advantage of the full power of deep protocol inspection, resulting in a marked
c
2013 SecurityMatters B.V.
11
www.secmatters.com
reduction in threat detection. Additional concerns and costs come from the fact that, each
time a system (e.g. a PLC) is recongured, updated or re-programmed, the rule-based
solution that protects it must be likewise (manually) updated.
3.3
Non-signature based
In order to enhance signature-based and ruled-based approaches, computer scientists have
been researching for more than a decade on automated techniques and algorithms to detect
misbehavior without requiring previous knowledge of it. These approaches are usually called
non-signature based (or anomaly-based). Today, most non-signature based solutions employ
articial intelligence algorithms, like neural networks, to rst learn what are the normal, legitimate \states" of an environment and afterwards raise an alert when they observe anomalous
network trac. These algorithms have been successfully applied to other elds in the past.
For instance, closed circuit television (CCTV) systems use articial intelligence algorithms for
recognizing a suspect among groups of people passing through border security.
When applied to network monitoring, non-signature based solutions come in two inherent
avors, depending on whether they carry out qualitative or quantitative analysis. There is
also a third, orthogonal category, called \sandbox" analysis. Finally, an innovative technology
has recently been proposed, called Deep Protocol Behavior Inspection (DPBI).
Qualitative analysis Qualitative analysis of network trac is based on the assumption that
malicious trac will look (signicantly) dierent from regular and legitimate trac. Solutions based on this approach observe network trac for a given amount of time (the learning
phase), build one or several models to describe what is expected to be regular trac, and
alert when the subsequently observed trac is \too dierent" from these models.
Pros: In theory, this approach could detect the most harmful and sophisticated attacks, as
for instance those aimed at executing arbitrary binary code on targeted assets by exploiting
unknown software vulnerabilities, deviating the underlying process by sending oending commands (or parameters).
Cons: Despite some promising results seen in initial studies, serious shortcomings became
obvious when this kind of solutions were deployed in real environments. When processing
real network trac, this approach falls short of identifying malicious behavior with sucient
accuracy [3]. The few qualitative solutions available are based on the assumption that it is
possible to seize network trac's intrinsic characteristics by using a generic model. Recent
studies [3, 6] prove not only that this assumption is often wrong, but also that these solutions
generate an extreme percentage of false alerts even when processing regular attack-free trac,
making them too inaccurate and far too expensive to manage. An additional issue is related to
tuning and customization. Tuning is essential in a dynamic environment such as a computer
network. Because qualitative solutions mostly work like \black boxes", leveraging complex
mathematical and statistical models, it is dicult, if not impossible, for users to tweak working
parameters to improve eectiveness.
Quantitative analysis (ow-based) Quantitative solutions build a model of network ows
by taking into consideration aggregate characteristics such as number of bytes exchanged
between network devices, number of new connections, etc., over a time window. When subsequently observed ows dier from the model, the system raises an alert. The built model is
typically automatically adjusted over time to incorporate changes in the network conguration.
Pros: The technology behind quantitative analysis is well established and has been revisited
over time to deal with new attacks. Quantitative solutions are usually easy to tune and tweak
c
2013 SecurityMatters B.V.
12
www.secmatters.com
for enhancing detection accuracy.
Cons: In practice, quantitative solutions can detect only misbehavior that generates spikes
in data volumes and network communications, such as a denial of service, horizontal and
vertical (port) scans, and brute-force attacks. This represents a severe limitation, because
most sophisticated attacks, or even erroneous commands issued by an operator, would remain
undetected, as they do not generate a spike in network ows.
Sandbox-based analysis This approach combines host- and network-based analysis techniques. The idea behind it is to capture \interesting" data from network trac, such as email
attachments, binary les, PDFs and Oce documents, and open or execute those les inside
a controlled environment (the sandbox). By enforcing certain controls over sensitive process,
data and conguration les (e.g. the Windows Registry), it is possible to detect when a
captured le could harm the targeted assets, for instance by dropping a botnet client.
Pros: This approach is very eective when it comes to detect advanced threats spreading through les, even when exploiting new and unknown software vulnerabilities. It does not
require signatures to work, although vendors typically include those to shorten detection time.
Cons: Sandbox analysis is only available for protecting the Windows OS family, and is strongly
oriented to malware detection. Hence, an attack carried out by using, for instance, an infected
laptop and aiming at exploiting software vulnerability in a PLC would go unnoticed, and so
would an erroneous command issued by an operator, because no malicious le is exchanged
in the process. Another issue is that attackers are aware of this monitoring technique and
started including obfuscation mechanisms to prevent the sandbox from detecting the threat.
For instance, a certain malware embedded in a pdf document could trigger only if the system
user was named \joe", or the organization's name was \ACME". This is just a simple example,
more complex and sophisticated obfuscation techniques have been seen in the wild.
Deep Protocol Behavior Inspection (DPBI) DPBI is an innovative non-signature based
technology developed by SecurityMatters, which combines the eectiveness of high accuracy
network whitelisting with the ease of use of a self-conguring system. DPBI performs the
most in-depth possible inspection of network communications, understanding both syntax and
semantics of ICS protocols. Network communications are analyzed down to the values exchanged by network devices, enabling the detection of any data-based threat (e.g. unknown
malware spread, illegitimate or unusual commands or values) on the most commonly used
ICS protocols, thereby providing protection for all network segments. In addition, the selfconguring technology enables to automatically dene the whitelisting rules by observing and
learning the normal trac within a network (communication patterns, protocols, message
types, and values exchanged). The result of this learning process is an accurate view of all
network activities, which allows to immediately detect any threat to the industrial process.
Pros: The combination of high accuracy network whitelisting and self-conguring technology
enables not only catch the most advanced attacks to ICS, but also to provide full situational
awareness and continuous monitoring of network and systems' behavior, allowing to detect,
e.g., system misconguration and conguration changes. The self-conguring technology
also allows to minimize re-conguration costs of the monitoring solution following changes
to the network and systems' conguration. Finally, DPBI is devised for \passive" network
monitoring, i.e. to observe network trac without blocking communications. This allows
for a higher level of detection without putting system performance and availability at risk by
erroneously blocking legitimate trac.
c
2013 SecurityMatters B.V.
13
www.secmatters.com
Cons: Despite providing protection from a wider range of threats and pinpointing their root
cause with high level of precision, DPBI does not directly \classify" the threats it detects, as a
signature-based solution would do. For instance, a signature-based solution could immediately
report an \SQL injection" if a byte pattern resembling such attack is observed, whereas DPBI
would simply report the suspicious sequence of bytes. This is because detection by DPBI is
not obtained by matching network trac with a database of known malicious payloads (i.e.
attack signatures), but rather by looking for abnormal network activities, and reporting what
parts of the communications were abnormal and how. Notice, however, that this detection
and classication capability of signature-based solutions is limited to attacks for which a
signature is devised, whereas DPBI enables detection of any kind of illegitimate behavior.
3.4
Technology comparison
Table 3.1 summarizes the detection capabilities of the dierent technologies available by
showing which of the threats presented in section 2.3 they would be able to detect.
Signature-based
Sandbox-based
Low accuracy
whitelisting
Medium accuracy
whitelisting
DPBI/High accuracy
whitelisting
Malicious/
careless
operators
Unexpected
conguration
shift
7
7
7
7
7
Spread of
malware
!
7
7
!
!
!
!
X
X
X
X
X
X
Targeted
attack
!
X
Table 3.1: Detection capabilities of the existing network monitoring technologies
Key: 7= No detection; != Partial detection; X= Full detection
Within the table, \Partial detection" means that the solution might detect the threat
depending on the way it is implemented. For example, signature-based solutions would detect
malware spread if the malware was exploiting known vulnerabilities or a signature for the
malware had already been devised.
c
2013 SecurityMatters B.V.
14
www.secmatters.com
4 Conclusions
ICS networks sit at the core of every industrial process. Recommendations, standards and
guidelines increasingly advocate the importance of implementing an adequate infrastructure
for continuously and accurately monitoring the activities in these networks, to protect them
from the growing number and range of cyberthreats they face.
Threats to a company's industrial processes can originate both in the corporate network
and in the Supervisory and Process Control Networks. Threats originating in the corporate
network are mainly represented by malware that attempts to spread to \lower" network
segments to compromise their components (e.g. SCADA servers and PLCs) and disrupt the
production process. The most notable and imminent threats, however, are those originating
in systems located within the Supervisory and Process Control Network, as these systems
have direct control of eld devices, sensors, and actuators. These threats include advanced
malware (e.g. downloaded from USB sticks or through malicious software/rmware patches
and updates ), direct data injection attacks and system misuse carried out by disgruntled
employees, contractors and vendors, but also unintentional mistakes by operators.
There are a few existing technologies that can be employed to monitor and protect ICS
networks. These technologies are complementary, as they are devised to detect dierent
types of threat. Signature-based solutions and sandbox analysis systems are designed to
protect corporate networks from known threats and unknown malware respectively, and to
prevent the spreading of those threats to other network segments, but leave Supervisory and
Process Control Networks mostly unprotected and susceptible to all the threats originating
inside those networks. Whitelisting solutions on the other hand are the ideal choice to protect
Supervisory and Process Control Networks, as they provide protection from any activity that
is not explicitly denoted as legitimate and authorized.
Figure 4.1 shows an example ICS network implementing a comprehensive monitoring infrastructure to achieve the best defense-in-depth. The infrastructure consists of:
1
A signature-based solution to stop known threats originating in the corporate network
1 http://www.informationweek.com/security/cybercrime/ame-hits-windows-update-7-keyfacts/240001490
c
2013 SecurityMatters B.V.
15
www.secmatters.com
A sandbox analysis system to detect unknown malware and prevent it from spreading
within the corporate network and to the Supervisory and Process Control Network
A whitelisting solution to protect OT components from advanced threats and detect
any undesired network operation
Flow-based analysis to protect eld devices from denial of service and brute-force attacks
A SIM/SIEM to integrate and correlate the input of all these solution and provide
monitoring personnel with a unied view of the current security status of an organization
Figure 4.1: A comprehensive monitoring infrastructure to protect all segments of an industrial
network
To conclude, we point out that, in order to be eective, any security solution needs to be
coupled with appropriate security procedures within the organization.
4.1
Acknowledgments
This work has been partially supported by the project \Cyber Security: a Fundamental Basis
for Smart Grids" (reference no. TKISG02020). We thank Christina Hofer and Fred Streeand
for their useful insights.
4.2
About SecurityMatters
SecurityMatters is a spin-o from the Dutch Technical Universities. Its research team works
with many dierent projects throughout the EU and USA and delivers game-changing network
monitoring and intrusion detection technology to make their customers more secure and in
control.
c
2013 SecurityMatters B.V.
16
www.secmatters.com
Some History When Damiano Bolzoni and Emmanuele Zambon left Italy and their jobs to
join the research group of Sandro Etalle in the Netherlands, they had the declared intention
to \create the most innovative and eective intrusion detection system of the world and build
an enterprise to bring it to the market". After years of research in which several dierent
technologies have been devised and tested, DPBI saw the light. It was a truly new, truly
eective way of realizing network monitoring and intrusion detection. SecurityMatters was
founded in 2008 and incorporated in December 2009. Under Emmanuele's lead, SecurityMatters development team built SilentDefense around DPBI. Version 1.0 was ready and rst
placed on the market in September 2011. In 2012, SecurityMatters won the SecurityMatters
wins the COMMIT Wetenschapsvalorisatieprijs.
4.3
About the Authors
Sandro Etalle (1965) Sandro Etalle was 19 when he founded his rst high-tech company.
Meanwhile, he nished his studies at the conservatory (ute) and in mathematics (cum laude).
In 1992, he left the business world to pursue an academic career. In 1995, he received a PhD
from the University of Amsterdam, and is now also full professor and head of the computer
security group at the Technical University of Eindhoven. He is one of the co-founders of
SecurityMatters. As scientist, Etalle wrote more than 100 articles cited more than 1500
times.
Cli H. Gregory (1948) Prior to his work in the private sector, Cli H. Gregory served
28 years in the U.S. Navy, including assignments developing cyber security programs. His
experience in Agile software product and process development, as well as business process
thought leadership give him a unique vision of how to improve the safety and security of IT
organizations. A published author and accomplished public speaker and trainer, Cli holds a
PhD in Math from the Dublin Institute of Technology (DIT) and a number of IT and security
certications. During his more than 35-year career in both government and private industry,
Cli has focused on Information Assurance and Systems Integrity.
Damiano Bolzoni (1981) Damiano Bolzoni graduated in 2005 in Computer Science from
the Ca' Foscari University of Venice with a thesis on anomaly-based network intrusion detection system, together with his fellow classmate Emmanuele Zambon. During his master
studies he worked for the Information Risk Management division of KPMG Italy. He then
moved to the Netherlands to pursue a PhD at the University of Twente. He was member of
the Italian Under 20 Athletics team and took part in several international competitions.
Emmanuele Zambon (1980) Emmanuele Zambon has been involved in computer security
since 2003. In 2005, he also had graduated in Computer Science from the Ca' Foscari
University of Venice with a thesis on Intrusion Detection Systems. During and after his
studies he has been employed in the Information Risk Management group of KPMG and
in Telecom Italia as a consultant, doing ethical hacking, network vulnerability/assessment,
and software development/performance tuning. In 2006, Zambon followed Bolzoni to the
Netherlands to work together on the development of a new technology (the core of which
now forms SilentDefense). Zambon received his PhD in IT Risk Management in 2011 from
the University of Twente. He is now working part-time as a post-doc at the University of
Twente, doing research on intrusion detection for industrial process automation networks,
and working part-time for SecurityMatters.
Daniel Trivellato (1983) Daniel Trivellato pursued his Master's degree in Computer Science
at the Free University of Bozen-Bolzano, Italy, graduating cum laude in 2007. In 2012, Daniel
received his PhD in computer security from the Technische Universiteit Eindhoven, where he
c
2013 SecurityMatters B.V.
17
www.secmatters.com
worked on the design and implementation of innovative access control solutions for dynamic,
distributed, heterogeneous systems. His work was carried out under the supervision of prof.
dr. Sandro Etalle. Since 2012, Daniel works as a project leader at SecurityMatters.
c
2013 SecurityMatters B.V.
18
www.secmatters.com
Bibliography
[1] National Technical Authority for Information Assurance. Technical risk assessment. HMG
IA Standard No. 1 Issue No: 3.51, UK Cabinet Oce, 2009.
[2] Gleb Gritsai, Alexander Timorin, Yury Goltsev, Roman Ilin, Sergey Gordeychik, and Anton
Karpin. Scada safety in numbers. Report to Congressional Requesters v1.1, Positive Technologies, 2012. http://www.ptsecurity.com/download/SCADA analytics english.pdf.
[3] D. Hadziosmanovic, L. Simionato, D. Bolzoni, E. Zambon, and S. Etalle. N-gram against
the machine: On the feasibility of the n-gram network analysis for binary protocols. In
Proceedings of the 15th International Symposium on Research in Attacks, Intrusions and
Defenses (RAID 2012), Amsterdam, The Netherlands, volume 7462 of Lecture Notes in
Computer Science, pages 354{373, 2012.
[4] United States Government Accountability Oce. Critical infrastructure protection. Report to Congressional Requesters 05-434, GAO, 2005. http://www.gao.gov/new.items/
d05434.pdf.
[5] Charles P. Peeger. Reections on the insider threat. In Salvatore J. Stolfo, Steven M.
Bellovin, Angelos D. Keromytis, Shlomo Hershkop, Sean W. Smith, and Sara Sinclair,
editors, Insider Attack and Cyber Security - Beyond the Hacker, volume 39 of Advances
in Information Security, pages 5{16. Springer, 2008.
[6] Y. Song, M. E. Locasto, A. Stavrou, A. D. Keromytis, and S. J. Stolfo. On the infeasibility
of modeling polymorphic shellcode. In Proceedings of the 14th ACM conference on
Computer and communications security, CCS '07, pages 541{551. ACM, 2007.
[7] K. Stouer, J. Falco, and K. Scarfone. Guide to industrial control systems (ics) security.
Special Publication 800-82, National Institute of Standards and Technology, 2011. http:
//csrc.nist.gov/publications/nistpubs/800-82/SP800-82-nal.pdf.
[8] Joseph Weiss. Protecting Industrial Control Systems from Electronic Threats. Momentum
Press, 2010.
c
2013 SecurityMatters B.V.
19
www.secmatters.com